What separates a good clip from a bad one, in numbers.
We graded 1 670 clips from 90 long videos on four axes — hook, flow, value, trend. Here's what the data says, including where it contradicts intuition. Important scope note: these are grades from our analysis engine, not views measured after posting.
The hook is the bottleneck — by far.
Across the four graded axes, the hook is the only one where top grades are scarce. 15% of clips earn an A on hook, versus 45.4% on trend. Finding a trending topic is three times easier than opening well.
Share of A-graded clips per axis (n = 1 670)
A weak hook doesn't degrade a clip: it collapses it.
You'd expect a smooth slope. There isn't one. From A to C, median overall score drifts from 88 to 76 — gentle erosion. From C to D, it drops from 76 to 41. The fall is brutal, across a single grade: there's no such thing as a 'mediocre hook' — there are hooks that hold and hooks that break.
Median overall score (out of 100) by hook grade
Duration predicts nothing.
It's the most repeated advice in the space — "keep clips short" — and our data doesn't support it. A-graded clips run 49s median, F-graded 43s. A six-second gap, in the opposite direction to intuition. Overall median is 46s. Stretching or trimming won't fix a weak opening.
Median duration in seconds, by hook grade
The moments you think are strong hook the worst.
This is the study's most useful result. The most common moment type — the classic 'strong moment', half our base at 838 clips — earns an A hook only 11.5% of the time. Meanwhile funny moments (50%) and strong opinions (39.5%) hook three to four times better, while being far rarer. What the ear registers as 'strong' isn't what stops the scroll.
Share of clips earning an A hook, by moment type
What these numbers say — and what they don't
Sample: 1 670 clips from 90 long videos (streams, podcasts, interviews, mostly French-language), sampled 2026-07-19. Each clip is graded A→F on four axes by the Nysos analysis engine.
Known limitation: these grades assess a moment's potential; they do not measure views obtained after posting. They describe what separates one clippable moment from another, not what performed on TikTok. A 1,670-clip sample is modest, and skews French-language.
Data freely reusable with a link to this page. Going further: paid clipping campaigns · the earnings calculator.
Questions about the study
What is this study based on?
On 1 670 clips from 90 long videos (streams, podcasts, interviews) processed through Nysos, each graded A to F on four axes: hook, flow, value, trend. Sampled 2026-07-19. These are grades from our analysis engine, not performance measured after posting.
Why is a strong hook so rare?
Because it depends on the first 3 seconds, not the topic. A moment can be gripping overall and still open on a soft transition, an open question, or a pronoun with no context. Our numbers: 15% As on hook versus 45.4% on trend — a trending topic is three times more common than a strong opening.
Should clips be short or long?
Duration does not predict hook quality. A-graded clips run 49s median, F-graded 43s — a 6-second gap with no clear hierarchy. Median duration across all grades is 46s. In other words: stretching or trimming a clip won't fix a weak opening.
Can I cite this data?
Yes, freely, with a link to this page. Please state the sample date and scope: analysis-engine grades on 1,670 clips, not view measurements.